Upload bvcc+nisqa+pstn+singmos+somos+tencent+tmhint-qi/alignnet/4337/config.yml with huggingface_hub
Browse files
bvcc+nisqa+pstn+singmos+somos+tencent+tmhint-qi/alignnet/4337/config.yml
ADDED
@@ -0,0 +1,60 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
additional_config: null
|
2 |
+
allow_cache: false
|
3 |
+
best_model_criterion:
|
4 |
+
key: utt_SRCC
|
5 |
+
order: highest
|
6 |
+
collater_type: NonIntrusiveCollater
|
7 |
+
config: conf/alignnet-wav2vec2.yaml
|
8 |
+
dataset_type: NonIntrusiveDataset
|
9 |
+
dev_csv_path: data/dev.csv
|
10 |
+
dev_samples_per_eval_loop: 1000
|
11 |
+
distributed: false
|
12 |
+
eval_and_save_interval_steps: 1000
|
13 |
+
grad_norm: 1.0
|
14 |
+
inference_mode: mean_listener
|
15 |
+
init_checkpoint: ''
|
16 |
+
keep_nbest_models: 5
|
17 |
+
listener_score_criterions: null
|
18 |
+
log_interval_steps: 100
|
19 |
+
mean_score_criterions:
|
20 |
+
- criterion_params:
|
21 |
+
masked_loss: false
|
22 |
+
order: 1
|
23 |
+
tau: 0.5
|
24 |
+
criterion_type: ScalarLoss
|
25 |
+
criterion_weight: 1.0
|
26 |
+
model_input: waveform
|
27 |
+
model_params:
|
28 |
+
decoder_dnn_dim: 64
|
29 |
+
domain_emb_dim: 128
|
30 |
+
output_type: scalar
|
31 |
+
range_clipping: true
|
32 |
+
s3prl_name: wav2vec2
|
33 |
+
ssl_model_layer_idx: -1
|
34 |
+
ssl_model_output_dim: 768
|
35 |
+
ssl_module: s3prl
|
36 |
+
use_domain_modeling: true
|
37 |
+
model_type: AlignNet
|
38 |
+
num_domains: 7
|
39 |
+
num_workers: 0
|
40 |
+
optimizer_params:
|
41 |
+
lr: 0.001
|
42 |
+
momentum: 0.9
|
43 |
+
optimizer_type: SGD
|
44 |
+
outdir: exp/alignnet-wav2vec2-4337
|
45 |
+
padding_mode: repetitive
|
46 |
+
patience: 20
|
47 |
+
pin_memory: true
|
48 |
+
rank: 0
|
49 |
+
resume: ''
|
50 |
+
sampling_rate: 16000
|
51 |
+
scheduler_type: null
|
52 |
+
seed: 4337
|
53 |
+
test_batch_size: 1
|
54 |
+
train_batch_size: 16
|
55 |
+
train_csv_path: data/train.csv
|
56 |
+
train_max_steps: 100000
|
57 |
+
trainer_type: NonIntrusiveEstimatorTrainer
|
58 |
+
verbose: 1
|
59 |
+
version: 0.1.0
|
60 |
+
wav_only: true
|